?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Staging: hv: remove OnChildDeviceRemove vmbus_driver callback
[zen-stable.git] / net / ipv4 / ip_options.c
blob1906fa35860c88a0919cd434268cbbcd679a3c39
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * The options processing module for ip.c
7 *
8 * Authors: A.N.Kuznetsov
9 *
10 */
11
12 #include <linux/capability.h>
13 #include <linux/module.h>
14 #include <linux/slab.h>
15 #include <linux/types.h>
16 #include <asm/uaccess.h>
17 #include <linux/skbuff.h>
18 #include <linux/ip.h>
19 #include <linux/icmp.h>
20 #include <linux/netdevice.h>
21 #include <linux/rtnetlink.h>
22 #include <net/sock.h>
23 #include <net/ip.h>
24 #include <net/icmp.h>
25 #include <net/route.h>
26 #include <net/cipso_ipv4.h>
27
28 /*
29 * Write options to IP header, record destination address to
30 * source route option, address of outgoing interface
31 * (we should already know it, so that this function is allowed be
32 * called only after routing decision) and timestamp,
33 * if we originate this datagram.
34 *
35 * daddr is real destination address, next hop is recorded in IP header.
36 * saddr is address of outgoing interface.
37 */
38
39 void ip_options_build(struct sk_buff * skb, struct ip_options * opt,
40 __be32 daddr, struct rtable *rt, int is_frag)
41 {
42 unsigned char *iph = skb_network_header(skb);
43
44 memcpy(&(IPCB(skb)->opt), opt, sizeof(struct ip_options));
45 memcpy(iph+sizeof(struct iphdr), opt->__data, opt->optlen);
46 opt = &(IPCB(skb)->opt);
47
48 if (opt->srr)
49 memcpy(iph+opt->srr+iph[opt->srr+1]-4, &daddr, 4);
50
51 if (!is_frag) {
52 if (opt->rr_needaddr)
53 ip_rt_get_source(iph+opt->rr+iph[opt->rr+2]-5, rt);
54 if (opt->ts_needaddr)
55 ip_rt_get_source(iph+opt->ts+iph[opt->ts+2]-9, rt);
56 if (opt->ts_needtime) {
57 struct timespec tv;
58 __be32 midtime;
59 getnstimeofday(&tv);
60 midtime = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC + tv.tv_nsec / NSEC_PER_MSEC);
61 memcpy(iph+opt->ts+iph[opt->ts+2]-5, &midtime, 4);
62 }
63 return;
64 }
65 if (opt->rr) {
66 memset(iph+opt->rr, IPOPT_NOP, iph[opt->rr+1]);
67 opt->rr = 0;
68 opt->rr_needaddr = 0;
69 }
70 if (opt->ts) {
71 memset(iph+opt->ts, IPOPT_NOP, iph[opt->ts+1]);
72 opt->ts = 0;
73 opt->ts_needaddr = opt->ts_needtime = 0;
74 }
75 }
76
77 /*
78 * Provided (sopt, skb) points to received options,
79 * build in dopt compiled option set appropriate for answering.
80 * i.e. invert SRR option, copy anothers,
81 * and grab room in RR/TS options.
82 *
83 * NOTE: dopt cannot point to skb.
84 */
85
86 int ip_options_echo(struct ip_options * dopt, struct sk_buff * skb)
87 {
88 struct ip_options *sopt;
89 unsigned char *sptr, *dptr;
90 int soffset, doffset;
91 int optlen;
92 __be32 daddr;
93
94 memset(dopt, 0, sizeof(struct ip_options));
95
96 sopt = &(IPCB(skb)->opt);
97
98 if (sopt->optlen == 0) {
99 dopt->optlen = 0;
100 return 0;
101 }
102
103 sptr = skb_network_header(skb);
104 dptr = dopt->__data;
105
106 daddr = skb_rtable(skb)->rt_spec_dst;
107
108 if (sopt->rr) {
109 optlen = sptr[sopt->rr+1];
110 soffset = sptr[sopt->rr+2];
111 dopt->rr = dopt->optlen + sizeof(struct iphdr);
112 memcpy(dptr, sptr+sopt->rr, optlen);
113 if (sopt->rr_needaddr && soffset <= optlen) {
114 if (soffset + 3 > optlen)
115 return -EINVAL;
116 dptr[2] = soffset + 4;
117 dopt->rr_needaddr = 1;
118 }
119 dptr += optlen;
120 dopt->optlen += optlen;
121 }
122 if (sopt->ts) {
123 optlen = sptr[sopt->ts+1];
124 soffset = sptr[sopt->ts+2];
125 dopt->ts = dopt->optlen + sizeof(struct iphdr);
126 memcpy(dptr, sptr+sopt->ts, optlen);
127 if (soffset <= optlen) {
128 if (sopt->ts_needaddr) {
129 if (soffset + 3 > optlen)
130 return -EINVAL;
131 dopt->ts_needaddr = 1;
132 soffset += 4;
133 }
134 if (sopt->ts_needtime) {
135 if (soffset + 3 > optlen)
136 return -EINVAL;
137 if ((dptr[3]&0xF) != IPOPT_TS_PRESPEC) {
138 dopt->ts_needtime = 1;
139 soffset += 4;
140 } else {
141 dopt->ts_needtime = 0;
142
143 if (soffset + 8 <= optlen) {
144 __be32 addr;
145
146 memcpy(&addr, sptr+soffset-1, 4);
147 if (inet_addr_type(dev_net(skb_dst(skb)->dev), addr) != RTN_LOCAL) {
148 dopt->ts_needtime = 1;
149 soffset += 8;
150 }
151 }
152 }
153 }
154 dptr[2] = soffset;
155 }
156 dptr += optlen;
157 dopt->optlen += optlen;
158 }
159 if (sopt->srr) {
160 unsigned char * start = sptr+sopt->srr;
161 __be32 faddr;
162
163 optlen = start[1];
164 soffset = start[2];
165 doffset = 0;
166 if (soffset > optlen)
167 soffset = optlen + 1;
168 soffset -= 4;
169 if (soffset > 3) {
170 memcpy(&faddr, &start[soffset-1], 4);
171 for (soffset-=4, doffset=4; soffset > 3; soffset-=4, doffset+=4)
172 memcpy(&dptr[doffset-1], &start[soffset-1], 4);
173 /*
174 * RFC1812 requires to fix illegal source routes.
175 */
176 if (memcmp(&ip_hdr(skb)->saddr,
177 &start[soffset + 3], 4) == 0)
178 doffset -= 4;
179 }
180 if (doffset > 3) {
181 memcpy(&start[doffset-1], &daddr, 4);
182 dopt->faddr = faddr;
183 dptr[0] = start[0];
184 dptr[1] = doffset+3;
185 dptr[2] = 4;
186 dptr += doffset+3;
187 dopt->srr = dopt->optlen + sizeof(struct iphdr);
188 dopt->optlen += doffset+3;
189 dopt->is_strictroute = sopt->is_strictroute;
190 }
191 }
192 if (sopt->cipso) {
193 optlen = sptr[sopt->cipso+1];
194 dopt->cipso = dopt->optlen+sizeof(struct iphdr);
195 memcpy(dptr, sptr+sopt->cipso, optlen);
196 dptr += optlen;
197 dopt->optlen += optlen;
198 }
199 while (dopt->optlen & 3) {
200 *dptr++ = IPOPT_END;
201 dopt->optlen++;
202 }
203 return 0;
204 }
205
206 /*
207 * Options "fragmenting", just fill options not
208 * allowed in fragments with NOOPs.
209 * Simple and stupid 8), but the most efficient way.
210 */
211
212 void ip_options_fragment(struct sk_buff * skb)
213 {
214 unsigned char *optptr = skb_network_header(skb) + sizeof(struct iphdr);
215 struct ip_options * opt = &(IPCB(skb)->opt);
216 int l = opt->optlen;
217 int optlen;
218
219 while (l > 0) {
220 switch (*optptr) {
221 case IPOPT_END:
222 return;
223 case IPOPT_NOOP:
224 l--;
225 optptr++;
226 continue;
227 }
228 optlen = optptr[1];
229 if (optlen<2 || optlen>l)
230 return;
231 if (!IPOPT_COPIED(*optptr))
232 memset(optptr, IPOPT_NOOP, optlen);
233 l -= optlen;
234 optptr += optlen;
235 }
236 opt->ts = 0;
237 opt->rr = 0;
238 opt->rr_needaddr = 0;
239 opt->ts_needaddr = 0;
240 opt->ts_needtime = 0;
241 }
242
243 /*
244 * Verify options and fill pointers in struct options.
245 * Caller should clear *opt, and set opt->data.
246 * If opt == NULL, then skb->data should point to IP header.
247 */
248
249 int ip_options_compile(struct net *net,
250 struct ip_options * opt, struct sk_buff * skb)
251 {
252 int l;
253 unsigned char * iph;
254 unsigned char * optptr;
255 int optlen;
256 unsigned char * pp_ptr = NULL;
257 struct rtable *rt = NULL;
258
259 if (skb != NULL) {
260 rt = skb_rtable(skb);
261 optptr = (unsigned char *)&(ip_hdr(skb)[1]);
262 } else
263 optptr = opt->__data;
264 iph = optptr - sizeof(struct iphdr);
265
266 for (l = opt->optlen; l > 0; ) {
267 switch (*optptr) {
268 case IPOPT_END:
269 for (optptr++, l--; l>0; optptr++, l--) {
270 if (*optptr != IPOPT_END) {
271 *optptr = IPOPT_END;
272 opt->is_changed = 1;
273 }
274 }
275 goto eol;
276 case IPOPT_NOOP:
277 l--;
278 optptr++;
279 continue;
280 }
281 optlen = optptr[1];
282 if (optlen<2 || optlen>l) {
283 pp_ptr = optptr;
284 goto error;
285 }
286 switch (*optptr) {
287 case IPOPT_SSRR:
288 case IPOPT_LSRR:
289 if (optlen < 3) {
290 pp_ptr = optptr + 1;
291 goto error;
292 }
293 if (optptr[2] < 4) {
294 pp_ptr = optptr + 2;
295 goto error;
296 }
297 /* NB: cf RFC-1812 5.2.4.1 */
298 if (opt->srr) {
299 pp_ptr = optptr;
300 goto error;
301 }
302 if (!skb) {
303 if (optptr[2] != 4 || optlen < 7 || ((optlen-3) & 3)) {
304 pp_ptr = optptr + 1;
305 goto error;
306 }
307 memcpy(&opt->faddr, &optptr[3], 4);
308 if (optlen > 7)
309 memmove(&optptr[3], &optptr[7], optlen-7);
310 }
311 opt->is_strictroute = (optptr[0] == IPOPT_SSRR);
312 opt->srr = optptr - iph;
313 break;
314 case IPOPT_RR:
315 if (opt->rr) {
316 pp_ptr = optptr;
317 goto error;
318 }
319 if (optlen < 3) {
320 pp_ptr = optptr + 1;
321 goto error;
322 }
323 if (optptr[2] < 4) {
324 pp_ptr = optptr + 2;
325 goto error;
326 }
327 if (optptr[2] <= optlen) {
328 if (optptr[2]+3 > optlen) {
329 pp_ptr = optptr + 2;
330 goto error;
331 }
332 if (skb) {
333 memcpy(&optptr[optptr[2]-1], &rt->rt_spec_dst, 4);
334 opt->is_changed = 1;
335 }
336 optptr[2] += 4;
337 opt->rr_needaddr = 1;
338 }
339 opt->rr = optptr - iph;
340 break;
341 case IPOPT_TIMESTAMP:
342 if (opt->ts) {
343 pp_ptr = optptr;
344 goto error;
345 }
346 if (optlen < 4) {
347 pp_ptr = optptr + 1;
348 goto error;
349 }
350 if (optptr[2] < 5) {
351 pp_ptr = optptr + 2;
352 goto error;
353 }
354 if (optptr[2] <= optlen) {
355 __be32 *timeptr = NULL;
356 if (optptr[2]+3 > optptr[1]) {
357 pp_ptr = optptr + 2;
358 goto error;
359 }
360 switch (optptr[3]&0xF) {
361 case IPOPT_TS_TSONLY:
362 opt->ts = optptr - iph;
363 if (skb)
364 timeptr = (__be32*)&optptr[optptr[2]-1];
365 opt->ts_needtime = 1;
366 optptr[2] += 4;
367 break;
368 case IPOPT_TS_TSANDADDR:
369 if (optptr[2]+7 > optptr[1]) {
370 pp_ptr = optptr + 2;
371 goto error;
372 }
373 opt->ts = optptr - iph;
374 if (skb) {
375 memcpy(&optptr[optptr[2]-1], &rt->rt_spec_dst, 4);
376 timeptr = (__be32*)&optptr[optptr[2]+3];
377 }
378 opt->ts_needaddr = 1;
379 opt->ts_needtime = 1;
380 optptr[2] += 8;
381 break;
382 case IPOPT_TS_PRESPEC:
383 if (optptr[2]+7 > optptr[1]) {
384 pp_ptr = optptr + 2;
385 goto error;
386 }
387 opt->ts = optptr - iph;
388 {
389 __be32 addr;
390 memcpy(&addr, &optptr[optptr[2]-1], 4);
391 if (inet_addr_type(net, addr) == RTN_UNICAST)
392 break;
393 if (skb)
394 timeptr = (__be32*)&optptr[optptr[2]+3];
395 }
396 opt->ts_needtime = 1;
397 optptr[2] += 8;
398 break;
399 default:
400 if (!skb && !capable(CAP_NET_RAW)) {
401 pp_ptr = optptr + 3;
402 goto error;
403 }
404 break;
405 }
406 if (timeptr) {
407 struct timespec tv;
408 __be32 midtime;
409 getnstimeofday(&tv);
410 midtime = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC + tv.tv_nsec / NSEC_PER_MSEC);
411 memcpy(timeptr, &midtime, sizeof(__be32));
412 opt->is_changed = 1;
413 }
414 } else {
415 unsigned overflow = optptr[3]>>4;
416 if (overflow == 15) {
417 pp_ptr = optptr + 3;
418 goto error;
419 }
420 opt->ts = optptr - iph;
421 if (skb) {
422 optptr[3] = (optptr[3]&0xF)|((overflow+1)<<4);
423 opt->is_changed = 1;
424 }
425 }
426 break;
427 case IPOPT_RA:
428 if (optlen < 4) {
429 pp_ptr = optptr + 1;
430 goto error;
431 }
432 if (optptr[2] == 0 && optptr[3] == 0)
433 opt->router_alert = optptr - iph;
434 break;
435 case IPOPT_CIPSO:
436 if ((!skb && !capable(CAP_NET_RAW)) || opt->cipso) {
437 pp_ptr = optptr;
438 goto error;
439 }
440 opt->cipso = optptr - iph;
441 if (cipso_v4_validate(skb, &optptr)) {
442 pp_ptr = optptr;
443 goto error;
444 }
445 break;
446 case IPOPT_SEC:
447 case IPOPT_SID:
448 default:
449 if (!skb && !capable(CAP_NET_RAW)) {
450 pp_ptr = optptr;
451 goto error;
452 }
453 break;
454 }
455 l -= optlen;
456 optptr += optlen;
457 }
458
459 eol:
460 if (!pp_ptr)
461 return 0;
462
463 error:
464 if (skb) {
465 icmp_send(skb, ICMP_PARAMETERPROB, 0, htonl((pp_ptr-iph)<<24));
466 }
467 return -EINVAL;
468 }
469 EXPORT_SYMBOL(ip_options_compile);
470
471 /*
472 * Undo all the changes done by ip_options_compile().
473 */
474
475 void ip_options_undo(struct ip_options * opt)
476 {
477 if (opt->srr) {
478 unsigned char * optptr = opt->__data+opt->srr-sizeof(struct iphdr);
479 memmove(optptr+7, optptr+3, optptr[1]-7);
480 memcpy(optptr+3, &opt->faddr, 4);
481 }
482 if (opt->rr_needaddr) {
483 unsigned char * optptr = opt->__data+opt->rr-sizeof(struct iphdr);
484 optptr[2] -= 4;
485 memset(&optptr[optptr[2]-1], 0, 4);
486 }
487 if (opt->ts) {
488 unsigned char * optptr = opt->__data+opt->ts-sizeof(struct iphdr);
489 if (opt->ts_needtime) {
490 optptr[2] -= 4;
491 memset(&optptr[optptr[2]-1], 0, 4);
492 if ((optptr[3]&0xF) == IPOPT_TS_PRESPEC)
493 optptr[2] -= 4;
494 }
495 if (opt->ts_needaddr) {
496 optptr[2] -= 4;
497 memset(&optptr[optptr[2]-1], 0, 4);
498 }
499 }
500 }
501
502 static struct ip_options *ip_options_get_alloc(const int optlen)
503 {
504 return kzalloc(sizeof(struct ip_options) + ((optlen + 3) & ~3),
505 GFP_KERNEL);
506 }
507
508 static int ip_options_get_finish(struct net *net, struct ip_options **optp,
509 struct ip_options *opt, int optlen)
510 {
511 while (optlen & 3)
512 opt->__data[optlen++] = IPOPT_END;
513 opt->optlen = optlen;
514 if (optlen && ip_options_compile(net, opt, NULL)) {
515 kfree(opt);
516 return -EINVAL;
517 }
518 kfree(*optp);
519 *optp = opt;
520 return 0;
521 }
522
523 int ip_options_get_from_user(struct net *net, struct ip_options **optp,
524 unsigned char __user *data, int optlen)
525 {
526 struct ip_options *opt = ip_options_get_alloc(optlen);
527
528 if (!opt)
529 return -ENOMEM;
530 if (optlen && copy_from_user(opt->__data, data, optlen)) {
531 kfree(opt);
532 return -EFAULT;
533 }
534 return ip_options_get_finish(net, optp, opt, optlen);
535 }
536
537 int ip_options_get(struct net *net, struct ip_options **optp,
538 unsigned char *data, int optlen)
539 {
540 struct ip_options *opt = ip_options_get_alloc(optlen);
541
542 if (!opt)
543 return -ENOMEM;
544 if (optlen)
545 memcpy(opt->__data, data, optlen);
546 return ip_options_get_finish(net, optp, opt, optlen);
547 }
548
549 void ip_forward_options(struct sk_buff *skb)
550 {
551 struct ip_options * opt = &(IPCB(skb)->opt);
552 unsigned char * optptr;
553 struct rtable *rt = skb_rtable(skb);
554 unsigned char *raw = skb_network_header(skb);
555
556 if (opt->rr_needaddr) {
557 optptr = (unsigned char *)raw + opt->rr;
558 ip_rt_get_source(&optptr[optptr[2]-5], rt);
559 opt->is_changed = 1;
560 }
561 if (opt->srr_is_hit) {
562 int srrptr, srrspace;
563
564 optptr = raw + opt->srr;
565
566 for ( srrptr=optptr[2], srrspace = optptr[1];
567 srrptr <= srrspace;
568 srrptr += 4
569 ) {
570 if (srrptr + 3 > srrspace)
571 break;
572 if (memcmp(&rt->rt_dst, &optptr[srrptr-1], 4) == 0)
573 break;
574 }
575 if (srrptr + 3 <= srrspace) {
576 opt->is_changed = 1;
577 ip_rt_get_source(&optptr[srrptr-1], rt);
578 ip_hdr(skb)->daddr = rt->rt_dst;
579 optptr[2] = srrptr+4;
580 } else if (net_ratelimit())
581 printk(KERN_CRIT "ip_forward(): Argh! Destination lost!\n");
582 if (opt->ts_needaddr) {
583 optptr = raw + opt->ts;
584 ip_rt_get_source(&optptr[optptr[2]-9], rt);
585 opt->is_changed = 1;
586 }
587 }
588 if (opt->is_changed) {
589 opt->is_changed = 0;
590 ip_send_check(ip_hdr(skb));
591 }
592 }
593
594 int ip_options_rcv_srr(struct sk_buff *skb)
595 {
596 struct ip_options *opt = &(IPCB(skb)->opt);
597 int srrspace, srrptr;
598 __be32 nexthop;
599 struct iphdr *iph = ip_hdr(skb);
600 unsigned char *optptr = skb_network_header(skb) + opt->srr;
601 struct rtable *rt = skb_rtable(skb);
602 struct rtable *rt2;
603 unsigned long orefdst;
604 int err;
605
606 if (!opt->srr)
607 return 0;
608
609 if (skb->pkt_type != PACKET_HOST)
610 return -EINVAL;
611 if (rt->rt_type == RTN_UNICAST) {
612 if (!opt->is_strictroute)
613 return 0;
614 icmp_send(skb, ICMP_PARAMETERPROB, 0, htonl(16<<24));
615 return -EINVAL;
616 }
617 if (rt->rt_type != RTN_LOCAL)
618 return -EINVAL;
619
620 for (srrptr=optptr[2], srrspace = optptr[1]; srrptr <= srrspace; srrptr += 4) {
621 if (srrptr + 3 > srrspace) {
622 icmp_send(skb, ICMP_PARAMETERPROB, 0, htonl((opt->srr+2)<<24));
623 return -EINVAL;
624 }
625 memcpy(&nexthop, &optptr[srrptr-1], 4);
626
627 orefdst = skb->_skb_refdst;
628 skb_dst_set(skb, NULL);
629 err = ip_route_input(skb, nexthop, iph->saddr, iph->tos, skb->dev);
630 rt2 = skb_rtable(skb);
631 if (err || (rt2->rt_type != RTN_UNICAST && rt2->rt_type != RTN_LOCAL)) {
632 skb_dst_drop(skb);
633 skb->_skb_refdst = orefdst;
634 return -EINVAL;
635 }
636 refdst_drop(orefdst);
637 if (rt2->rt_type != RTN_LOCAL)
638 break;
639 /* Superfast 8) loopback forward */
640 memcpy(&iph->daddr, &optptr[srrptr-1], 4);
641 opt->is_changed = 1;
642 }
643 if (srrptr <= srrspace) {
644 opt->srr_is_hit = 1;
645 opt->is_changed = 1;
646 }
647 return 0;
648 }
649 EXPORT_SYMBOL(ip_options_rcv_srr);